The dataset contains images of Diabetic Retinopathy stages and each stage has its respective folder. The images were taken from three different sources and were merged together.
- APTOS2015 Competition (Kaggle) https://www.kaggle.com/c/diabetic-retinopathy-detection
- APTOS2019 Competition (Kaggle) https://www.kaggle.com/c/aptos2019-blindness-detection
- Messidor-2 Dataset https://www.adcis.net/en/third-party/messidor2/
Each images were first preprocessed using OpenCV in Python based Ben Graham implementation in a APTOS2015 Kaggle competition where he achieved 1st position. In the preprocessing step,
- First the image is blurred using Gaussian Filter.
- Followed by element-wise subtraction.
Then each image were resized to 256x256x3 and split into training and test sets.